Skip to content
Computer vision

NorthCUA-Pro

397B MoE-агент для управления компьютером

2 min read

NorthCUA-Pro — крупнейшая модель семейства NorthCUA от beipo-ai: 397B параметров всего и 17B активных, mixture-of-experts на базе Qwen3.5. Она принимает запросы, скриншоты и историю взаимодействия, а выдаёт рассуждения, ответы и tool calls для desktop- или браузерного runtime. Для computer-use нужен официальный NorthCUA Harness; координаты GUI строго нормализованы в normalized_1000 (0–1000), а перевод в пиксели выполняет Harness.

Ключевые особенности

  • Визуальное grounding и мультимодальные наблюдения для desktop-приложений и веб-сценариев.
  • Обучение на длинных траекториях: зависимые шаги, промежуточные артефакты и проверка результата.
  • Семейство: Nano 9B dense, Mini 35B total / 3B active MoE и Pro 397B total / 17B active MoE.
  • Открытые веса под Apache-2.0; в репозитории Safetensors, tokenizer, image processor и chat template.
  • Контекст 262 144 токена; 122 шарда, около 793.63 ГБ на диске.

Пример

import torch
from transformers import AutoProcessor, Qwen3_5MoeForConditionalGeneration
model_id = 'beipo-ai/NorthCUA-Pro'
processor = AutoProcessor.from_pretrained(model_id)
model = Qwen3_5MoeForConditionalGeneration.from_pretrained(model_id, dtype=torch.bfloat16, device_map='auto')
ПодробнееПо отчёту: ScreenSpot-Pro — 93.30% accuracy (SFT), OSWorld-G — 90.96%, OSWorld-Verified — 88.64% success (RL), WebArena-Verified — 73.89% (RL). Это исследовательские результаты, не новые замеры именно этих весов.
Для весов нужно около 793.63 ГБ на диске; оценок VRAM в карточке нет, нужен достаточный суммарный объём памяти ускорителей.
Действия и заявленные результаты могут быть неверными: проверяйте важные изменения и изолируйте окружение.
Примеры
Discuss Telegram 12 446

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • BF16weights 739 GB · ready files814 GB
  • FP8 / INT8recommendedweights 370 GB407 GB
  • 4-bitweights 208 GB229 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

No single instance holds the whole model: it needs RAM offload or a cluster.

Your own GPU: Won't fit any home GPU — renting is cheaper.

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # beipo-ai/NorthCUA-Pro
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download beipo-ai/NorthCUA-Pro --local-dir ~/models/NorthCUA-Pro
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to NorthCUA-Pro
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of NorthCUA-Pro (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face

About the model

beipo-ai/NorthCUA-Pro
Author
beipo-ai
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

44612

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Computer vision

Locus-2B

Locus-2B
Computer vision

LFM3-VLD-2B

LFM3-VLD-2B